bceloss 和CrossEntropyLoss

 

#!/usr/bin/env python
# -*- coding: utf-8 -*-
import time

import torch

import torch

v = 0.5  # 1-0.0001
v1 = v - 0.01
a = torch.FloatTensor([[v, v1, v],[v, v1, v]])
b = torch.FloatTensor([[0, 0, 0],[1, 1, 1]])

loss_fn = torch.nn.BCELoss()  # reduce=False, size_average=False)

x = loss_fn(a, b).item()
print(x)

a=a.view(-1,2)
print(a)
b=b.long()
cross= torch.nn.CrossEntropyLoss(reduction='sum')

aaa= cross(a, b)
print(aaa)



    # tensor([[0.1933, 0.1425, 0.8572, 0.0224, 0.3811],
    #         [0.6134, 0.9766, 0.6086, 0.0163, 0.1514]])
    # tensor([[2, 4, 0, 1, 3],
    #         [1, 0, 2, 4, 3]])
    # tensor([[2, 3, 0, 4, 1],
    #         [1, 0, 2, 4, 3]])
    #最小的索引在第2个位置,次小的索引在第3位
    # tensor([[False, False, True, False, False],
    #         [True, True, True, False, False]])
    # tensor([0.8572, 0.6134, 0.9766, 0.6086])